method
active
method:many-shot-steering-vector-computationMany-shot steering vector computation
Exploratory experiment computing vectors over multiple shots.
Neighborhood — ranked by edge-count
Papers (1)
paper
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- A method for modifying model behavior by adding perturbation vectors to activations, used here to try to reduce eval awareness.
- Steering vectors computed over multiple shots and norm-normalized have weaker effect than 0-shot vectorsfinding0.793Counter-intuitive result.
- Constructing steering vectors from the difference of mean activations on positive and negative examples, for comparison.
- Why do steering vectors computed over multiple shots have a weaker effect than single-query vectors?question0.786Left to future work.
- Used to extract trait and style directions.
- Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.
- Translates steering vector via weighted average of shared decoder vectors.
- Method for computing steering vectors as mean activation differences between reflection levels at a given layer.